LDLMIn construction

All models, with their blocks →

Size
306M trained
Published
Meshchaninov et al. · May 2026
Line
Diffusion language model
View
Architecture
Source

LDLM @ e18c42a, not yet public · arXiv:2605.07933v1 · OpenWebText · 1024 · checkpoint, tokenizer and saved statistics not verifiedHow to read

Time embedding

TrainingGeneration

Sampler › Denoiser › Time embedding

Expansion

×

learned

Map the 768 encoding features to 1,536.

e↑=etW1+b1{\htmlData{light=time-up}{e^{\uparrow}}}={\htmlData{light=sinusoid}{e_t}}W_{1}+b_{1}
in
ete_t [B batch, 768 features]
out
e↑e^{\uparrow} [B batch, 1536 features]
parameters
1,181,184
Parameters

1,181,184 parameters owned by this component

Exact per-instance accounting
ComponentCalculationParameters
weights + bias768×1,536+1,536768 \times 1{,}536 + 1{,}5361,181,184
Per-instance total1,181,184

Across 1024 shared calls: 1,181,184×1=1,181,1841{,}181{,}184 \times 1 = 1{,}181{,}184 parameters.

Derived from the pinned implementation’s constructors and configuration. The checkpoint’s tensors have not been counted. 306,011,162 of these parameters are trained; frozen GPT-2 and the decoder’s mask vector, 124,442,112, are not.

Sources2

Direct evidence

The denoiser owns the time embedding’s two affine layers, one rotary table for all blocks, an affine projection of the concatenated input, and the stack.

The implementation, pinned at revision e18c42acdfb7, not yet public

Direct evidence

The time embedding passes its MLP and a SiLU; the noised latents are concatenated with the previous clean estimate, zeros when there is none, and projected; the stack runs in bfloat16.

The implementation, pinned at revision e18c42acdfb7, not yet public